Use Claude Code with a Screen Reader

Anthropic's coding agent Claude Code normally draws its terminal interface with boxes, animated spinners and text that redraws in place. That is hard to follow with a screen reader, which reads text in order. The accessibility documentation describes a screen reader mode that prints flat, scrolling text instead. It drops box-drawing characters and cues carried only by color, shows spinners as static text and rewrites tables as Header: value sentences. Each message begins with a spoken label such as you:, claude:, tool: or Permission Required:. Because everything stays in the terminal's scrollback, a user can search for those labels to jump between parts of the conversation.

The biggest change is in approving the agent's work. Menus that normally take arrow keys, including permission prompts, become numbered lists answered by typing a number, and yes-or-no prompts take a typed y or n. A proposed file edit is read as plain lines marked + for additions and - for removals, so the user hears the change before approving it. The terminal bell rings when Claude finishes a reply, when a prompt needs an answer, or when a tool that ran longer than 5 seconds finishes. That means the user does not have to keep re-reading the transcript to see whether anything is waiting.

ClaudeDevs announced the mode with the claude --ax-screen-reader flag. The documentation adds an environment variable and a user setting for keeping it on, and says Claude Code ignores requests for the mode in a project's own settings files. The mode does not switch on when a screen reader is detected. Some gaps remain. Attaching to a background session opens a screen with no native scrollback, switching to plan mode with the /plan command is not announced, and costs are reported only when the session exits. The VS Code extension's chat panel does not need the mode, and from version 2.1.236 it announces conversation activity to a screen reader without any setting.